Search the Atlas

Search risks, controls, and glossary terms

HighOpacityRealized

Black-box unexplainable output

Transparency & Explainability

Description

Generated decisions/answers cannot be explained, preventing review, recourse, or regulatory justification.

Example scenario

A declined customer cannot be given a meaningful, accurate reason for an AI-assisted decision.

Real-world evidenceRealized

Multiple confirmed production cases exist of AI systems deployed without disclosing their non-human nature to users, including customer service chatbots and companion AI platforms. The FTC has issued guidance and taken action on deceptive AI identity practices, and the EU AI Act codifies this as a legal obligation precisely because undisclosed AI interaction had already become a documented consumer harm.

Primary mitigations

  • Rationale generation with citations
  • retrieval-grounded explanations
  • decision logging
  • XAI tooling.

Detection signals

Explanation-coverage metric; rationale-quality review.

Mitigating controls

6
Non-agentic controls

Related risks in Transparency & Explainability